Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
『영미연구』제39집 (2017): 179-200
Sentiment Analysis Utilizing Modal
Expressions*1)
Jang Ji-won·Kim Jee-Eun
[Abstract]
This research presents a classification of English modal expressions which affect
polarity values in sentiment analysis. Modality, commonly encountered in sentences,
conveys various semantic attitudes of a speaker. Because of its special characteristics,
the meaning of a modalized sentence is challenging to be correctly interpreted. The
sentiment expressed with modal expressions is also difficult to evaluate. In order to
increase the accuracy in evaluating sentiment values, modal expressions are collected
from movie reviews and analyzed with selected linguistic cues to classify. The
classification includes the expressions, its contextual components to co-occur, and the
resulting sentiment value. Since it also lists up the classes which have not been dealt
with in previous studies, it is expected to contribute performance improvement of a
sentiment analysis system.
Key Words: sentiment analysis, subjectivity, modality, modal verb, polarity value
* This work was supported by Hankuk University of Foreign Studies Research Fund of 2016.
180 영미연구 제39집
I. Introduction
Fast growth in internet access has resulted in mass production of users’ opinions on
various subjects such as products, services, politics, entertainments and so on. Such
user-generated contents are considered valuable for understanding and monitoring
public opinions since they convey people’s sentiments. Sentiment analysis categorizes
text with positive, negative or neutral values. In order to interpret and evaluate
sentiments, subjectivity identification has to be preceded; opinionated sentences need
to be distinguished from factual ones in documents. Both subjectivity and sentiment
can be identified by utilizing various linguistics cues with which many researches
have been done. Despite a large volume of literature on the subject, very few studies
have been performed using modality which not only affects the interpretation of
sentiments, but also triggers to identify opinions.
Modality is a category of linguistic meaning which commonly encountered in
English sentences utilizing modal verbs and also presents people’s attitudes towards
what they are saying. Since people express various semantic attitude utilizing
modality, such as possibility, certainty, necessity, desire, and so on, the sentences
including those meanings become the candidates for opinions. Accordingly, modality
affects the evaluation of sentiments; eliminating sentiment information and
intensifying or diminishing the degree of the sentiments.
The modality in linguistics is classified into several categories such as deontic,
epistemic, etc. and the number of categories varies depending on the researchers.
Even though such classification is crucial for figuring out modal meanings included
in sentences, it has to be revised for sentiment analysis because linguistic cues are
not explicitly represented to identify a modal class. In order to produce higher
accuracy than the ones suggested by previous studies, this paper proposes a linguistic
Sentiment Analysis Utilizing Modal Expressions 181
approach for identifying the value of sentiment expressed in English sentences with
modality. The focus of the approach is a classification of modal verbs which affects
the evaluation of sentiments.
In the following section, previous studies on sentiment analysis are introduced and
briefly reviewed. Section 3 describes linguistic modality and section 4 discusses a
classification of English modal expressions which influence sentiment values.
Finally, section 5 concludes the paper, summarizing the research and suggesting
future research works.
II. Related Studies
Deciding subjectivity is a prerequisite to evaluate the polarity of sentiments. Wilson
et al. (2005) introduces OpinionFinder, a system to identify and extract subjective
information in documents. OpinionFinder recognizes subjective sentences in
documents by marking the source of the subjectivity and using sentiment words as
clues. This system decides the subjectivity of the sentences in documents by marking
either subjective or objective.
Table 1. A brief summary of subjectivity marking in OpinionFinder
Input Subjectivity
Cigarettes are bad for you Objective
Cigarettes must be bad for you Subjective
Cigarettes might be bad for you subjective
As illustrated in Table 1, modality is used as a feature to identify subjectivity. The
182 영미연구 제39집
system identifies the sentences with a modal auxiliary verb as subjective while the
sentences without a modal verb as objective.
Table 2. A brief summary of polarity marking in OpinionFinder
Input Polarity
Cigarettes are not bad for you 409_412 negative
It does not evaluate the sentiment value of the sentence, but just locates a
sentiment word with its polarity value as illustrated in Table 2. It just provides the
location of the sentiment word, bad, marking its polarity value as negative. It seems
it has to be modified to function as a full-fledged system.
Some researches introduce more advanced sentiment analysis systems by providing
a wider range of scales of polarity values. Wiebe et al. (2004) creates a sentiment
lexicon with five scales; strong positive, weak positive, neutral, weak negative, and
strong negative. Similarly, Stanford University also provides five sentiment classes
(Socher et al., 2013). Other researches have implemented a finer-grained scale of
polarity values. Taboada et al. (2011) builds a sentiment lexicon with a ten-point
scale, from –5 to +5. A sentiment analysis system provided by Natural Language
Toolkit (NLTK) also provides more detailed sentiment scales; it is ranged from 0.1
to 1.0 for both positive and negative values respectively. However, complex-scaled
polarity values are hard to determine since the degree of intensity is not absolute but
relative.
Some sentiment analysis studies make use of particular syntactic patterns as a
linguistic cue to evaluate sentiment. Jindal and Liu (2006) and Ganapathibhotla and
Liu (2008) analyze sentences with comparatives. They identify comparative
expressions to determine the sentiment. Narayanan et al. (2009) conduct sentiment
Sentiment Analysis Utilizing Modal Expressions 183
analysis with conditional sentences. Only few researches have performed on modality
in the field of sentiment analysis. Liu et al. (2014) and Liu (2015) provide a
classification of modal verbs according to their polarity value. For example, they
interpret a sentence including a dynamic modal verb can as positive without
considering the context. However, can is ambiguous in that it also functions as
deontic and there is no linguistic clue to distinguish dynamic from deontic. This type
of blind-folded generalization cannot be applied to the evaluation of sentiment.
A sentiment analysis system provided by Stanford University (Socher et al., 2013)
calculates sentiment by summing up polarity values of a phrase level in comparison
to the other systems which utilizes words. However, it seems that this system does
not take into account modality as illustrated in Figure 1.
Figure 1. An example of sentiment analysis using the model
of Stanford University
The sentiment of the sentence should not be positive. However, it is incorrectly
evaluated as positive since it seems to focus on the sentiment words while ignoring
the modal auxiliary verb, should.
Another sentiment analysis system provided by NLTK determines polarity values
in documents. This model also does not consider a modal verb when evaluating
184 영미연구 제39집
sentiment as illustrated in Figure 2 and 3.
Figure 2. Sentiment analysis of a sentence without a modal verb by NLTK
Figure 3. Sentiment analysis of a sentence with a modal verb by NLTK
The sentences in Figure 2 and 3 have the identical structure except the modal
verb, would, which should result in different interpretation. However, the results of
the evaluation are the same; positive 0.3 and negative 0.7. This confirms that
modality has not been interpreted properly.
Sentiment Analysis Utilizing Modal Expressions 185
III. Linguistic Analysis of English Modality
Modality is a semantic grammatical category expressing semantic notions such as
possibility, probability, necessity, obligation, permission, ability, or intention, which
represent a speaker’s attitudes and semantic orientations toward an entity. Since it
adds a broad range of semantic nuances to an utterance, it is crucial to identify its
appropriate meaning within the context. This results in various researches on
modality in linguistics. Among the researches, this study has adopted a linguistic
model introduced by Palmer (2001). Modality in English is realized through mood
and a modal system.
3.1. English Mood
Mood is a grammatical category for signaling modality. Mood is categorized into
indicative, imperative, and subjunctive. Indicative mood is used to indicate and state
a fact. It is encoded by declarative and interrogative sentences. Imperative mood
expresses command, request, exhortation, or advice representing desirable states
which have not happened yet.
(1) Movie making usually takes a few months.
(2) I hate the movie released last week.
(3) Did you like the movie?
(4) Stop watching the movie!
Some declarative sentences do not carry any sentiment as in sentence (1) while
other sentences such as sentence (2) express a speaker’s opinion. Interrogative
186 영미연구 제39집
sentences hardly express a speaker’s opinion since it is a process of seeking
information rather than stating a fact as shown in sentence (3). Similar to
interrogative mood, imperative mood does not deliver a speaker’s sentiment as in (4).
Subjunctive mood expresses irrealis such as wish, possibility, judgment, obligation
or necessity, which has not yet been realized. Subjunctive mood expresses a
speaker’s sentiment although it represents counterfactuality. English subjunctive is
realized utilizing if clause with were, should/would/could + have + past participle
(pp), and so on.
(5) It would be nice if the screen were a little bigger.
(6) The director should have studied American culture a little closer.
Sentence (5) expresses the speaker’s negative sentiment towards the movie screen,
utilizing subjunctive mood. The movie screen is not big enough to satisfy the speaker
at the time of uttering the sentence which conveys a desire for the screen to be
bigger. Similarly, sentence (6) expresses the speaker’s negative sentiment on the
director while using should + have + pp.
3.2 English Modal System
English modal system is constructed using modal auxiliary verbs. By employing
different auxiliaries, three types of modal meanings are conveyed; epistemic, dynamic
and deontic.
(7) John may/will/can be at the theater to watch The Dark Night.
(8) He must be a famous director.
Sentiment Analysis Utilizing Modal Expressions 187
Epistemic modality expresses a speaker’s judgments on factual status as shown in
sentences (7) and (8). Epistemic modality is encoded using modal auxiliary verbs
including may, might, must, can, could and will to deliver various modal meanings
such as assumptive, speculative, potential, subjunctive and so on.
(9) John may/can book a movie ticket.
(10) John must/should book a movie ticket.
Deontic modality indicates obligation or permission by an external force. A list of
modal auxiliaries, may, can, must and should, expresses deontic modality as
illustrated in sentences (9) and (10). May/can in (9) encodes permission and
must/should in (10) expresses obligation.
(11) John can write a screenplay.
(12) John will finish editing the movie by tomorrow.
Dynamic modality expresses a speaker’s ability or willingness to do something.
Can is adopted to encode ability as in sentences (11) and will is used to mean
willingness of the subject of the sentence as presented in (12).
As illustrated in the example sentences above, modality conveys a speaker’s
sentiment. Proper interpretation of modality is necessary to identify accurate
sentiment values. It naturally has to be dealt with in sentiment analysis to produce
better results. Few researches analyze sentiment utilizing modality because linguistic
modality is categorized according to a meaning that each class expresses. Modal
auxiliary verbs present an ambiguity when it is used in a sentence to convey a modal
meaning. For example, can expresses multiple modal meanings including epistemic,
deontic, and dynamic. The class of the modality can only be identified by
188 영미연구 제39집
interpreting the meaning of the sentence. In sentiment analysis, however,
disambiguation of modal meaning requires explicit linguistic patterns which only few
researches have suggested.
IV. Classification of Modal Expressions in
Sentiment Analysis
To provide a classification of modal expressions for sentiment analysis, eight English
modal verbs are selected for this research; can, could, will, would, might, may,
should, and must. The total number of 13,378 sentences with those verbs are
collected from two different sources; 12,863 from the movie review data by Pang
and Lee (2004) and 515 New York Times movie reviews by Kim (2013). The
sentences are analyzed using the sentiment lexicon of Hu and Liu (2004) in order to
determine the polarity values of words. The sentences are examined to check how
modal verbs affect sentiment and how differently each modal verb behaves from one
another. As a result, modal verbs are found out to assign, shift or remove a polarity
value according to a particular pattern occurring in a sentence. The result are
described in comparison to the classification suggested by Liu et al. (2014) and Liu
(2015).
4.1. Modal Verbs with a Positive/Negative Word
A modal verb changes the polarity value when it occurs with a sentiment word,
either positive or negative. Liu (2015) claims that can or could assigns a positive
value when it is used to express ability as shown in sentences (13a) ~ (15a).
Sentiment Analysis Utilizing Modal Expressions 189
(13a) I can count on Apple.1) (+)2)
(14a) This device can deal with the connection problem. (+)
(15a) This phone can do speed dialing but my previous phone cannot. (+)
(13b) I count on Apple. (+)
(14b) This device deals with the connection problem. (+)
(15b) This phone does speed dialing but my previous phone does not. (+)
(16) A movie trailer can/could spoil the entire movie. (−)
Sentences (13a) ~ (15a) are, however, still positive without the modal verb. They
are interpreted as positive because of the underlined sentiment words as confirmed in
(13b) ~ (15b). Sentence (16) includes can/could and a negative sentiment word spoil,
which is interpreted as negative. This example contradicts what Liu (2015) claims; a
sentence with can is interpreted as positive. From the list of examples presented
above, the polarity value of a sentence with can/could is determined by a sentiment
word, not the modal verb itself.
Another set of modal verbs will/would affect the evaluation of sentiment in the
same way to can/could. Liu (2015), however, fails to provide a clear explanation
how will influences sentiment while he suggests that would occurring with a positive
sentiment word assigns negative sentiment as listed in sentences (17) and (18).
(17) I would have loved this product. (−)
(18) It would have been a good car. (−)
The sentiment value of (17) and (18) is evaluated as negative because of would
followed by have + pp. This pattern should be treated separately from would alone
since they assign different polarity values respectively.
190 영미연구 제39집
May and might have not been dealt with from previous studies. They behave in
the same way to can and could in assigning a polarity value. Sentence (19) including
a positive word recommend is interpreted towards positive, and sentence (20) with a
negative word rip to shreds towards negative.
(19) I may/might recommend everyone to watch this movie. (+)
(20) He may/might rip the screenplay to shreds. (−)
Modal verbs, should and must, are complicated to generalize their patterns. When
they occur with a positive sentiment word, the polarity value of the sentence varies
as listed in sentences (21) ~ (23).
(21) He should/must recommend everyone to watch this movie. (+)
(22) The actor should/must impress the audience. (0)
(23) The movie theater should/must be clean. (0/−)
Sentence (21) expresses positive sentiment when it contains a positive sentiment
word recommend while sentence (22) is evaluated as neutral even though its
components are identical as (21). The difference between the two sentences is the
type of a verb, whether to be a psych verb. Sentence (23) is ambiguous in that it can
be interpreted as either neutral or negative depending on the current state of the
subject.
A sentence with a negative sentiment word is not discussed with the reasons. It
does not appear in the data and the resulted meaning seems awkward because the
modal verbs express the expectation of the speaker. Since generalization of the
patterns for should and must is hard to draw with the current research, it has to be
examined further with more data.
Sentiment Analysis Utilizing Modal Expressions 191
4.2. Modal Verbs with a Negation Word NOT
The modal verbs changes polarity values when they occur with a negation word such
as not. Can and could present interesting behaviors with a negation word in a
sentence as shown in sentences (24) and (25).
(24) I can/could not count on Apple. (−)
(25) The movie trailer can/could not spoil the entire movie. (0/diminished −)
They reverse the polarity when it is followed by a positive sentiment word as
displayed in sentence (24). They, however, neutralize the polarity value when a
negation word and a negative sentiment word occur in the sentence as shown in (25).
The negation word occurring with a negative sentiment word does not reverse the
sentiment, rather it cancels out the negative sentiment in a system with the polarity
range of –1 to 1. It may diminish the sentiment with a finer-grained scale.
A list of modal verbs including will, would, may and might, behave in the same
way to can and could when they occur with a negation word as shown in sentences
(26) and (27).
(26) I will/would/may/might not recommend everyone to watch this movie. (−)
(27) He will/would/may/might not rip the screenplay to shreds.
(0/diminished −)
The negation word reverses the polarity value when used with a positive word and
neutralizes or diminishes the degree of sentiment when occurred with a negative
sentiment word. However, there are subtle differences in the degree of the resulted
sentiment depending on each modal verb.
192 영미연구 제39집
Liu (2015) generalizes that the sentences with should express negative sentiment
with a negation word by presenting sentences (28) and (29). However, the usage of
should is not so simple to be generalized as Liu (2015) argues. No clear patterns for
should determine the sentiment value to be positive or negative as shown in
sentences (30) and (31).
(28) They should not make the screen so big. (−)
(29) Kids should not act foolish. (−)
(30) Jumping into a pond should/must not be dangerous. (0)
(31) I should/must not recommend everyone to watch this movie. (−)
Another modal verb must has not been considered in the study by Liu (2015). It
influences the sentiment value in the same way as should works as shown in (30)
and (31).
4.3 Modal Verbs with Have + Past Participle
Modal verbs affect the polarity value of a sentence when they occur with a syntactic
pattern, have + pp. The research by Liu et al. (2014) and Liu (2015) have not hardly
discussed the sentiment shift caused by the modal verbs used with this particular
pattern. A list of can/may/might/must diminishes or neutralizes the degree of
sentiment when used with have + pp as in sentence (32).
(32) The psychological elements can/may/might/must have made this film
interesting. (0/diminished +)
(33) The psychological elements can/may/might/must have made this film
disgusting. (0/diminished −)
Sentiment Analysis Utilizing Modal Expressions 193
Since the modal verbs with have + pp encode speculative/assumptive/deductive
meaning, they naturally diminish or neutralize the sentiment as in (32) and (33).
Because of its nature, each modal verb creates subtle differences in the degree of the
resulted sentiment.
Another list of should/would/could + have + pp expresses hypothetical or
subjunctive meaning. When they occur with a sentiment word, they behave
differently from the rest of the modal verbs. The polarity value becomes negative
although the sentiment word is positive. Since the pattern, should/would/could + have
+ pp, is known to represent desire or regretful emotion, the sentences including this
pattern express negative sentiment, as in sentences (34) and (35).
(34) The theater should/would/could have bought a big screen. (−)
(35) ?The theater should/would/could have made a screen so small. (−)
Sentence (35) sounds awkward when it is identical to (34) except the sentiment
word being negative. Its sentiment remains negative but it is semantically anomalous
considering the domain of the data, namely movie reviews.3)
In comparison to the rest of the modal verbs, a sentence with this pattern does not
change its sentiment value even when it is negated. Its sentiment value remains the
same, being negative as shown in sentences (36) and (37).
(36) The entrapment scene should/would/could not have been boring. (−)
(37) ?The entrapment scene should/would/could not have been exciting. (−)
Similar to sentence (35) whose resulted meaning is awkward when the pattern is
used with a negative sentiment word, sentence (37) also sounds semantically odd
when occurred with a negation word and a positive sentiment word. Nevertheless, its
194 영미연구 제39집
polarity value stays negative. Again, the degree of the sentiment varies depending on
each modal verb to be included in the sentence. When would or could is employed
in a sentence, an assumptive or speculative meaning can be read from a sentence.
Further researches on would and could with have + pp are required.
Table 3. A sentiment classification of modal expressions
Modal expressions Evaluated Sentiment
Can/could/will/would/may/might + a positive sentiment word
Positive or diminished positive
Can/could/will/would/may/might + a negative sentiment word
Negative or diminished negative
Should/must+ a positive/negative sentiment word
Positive, negative, or neutral
Can/could/will/would/may/might + NOT + a positive sentiment word
Negative
Can/could/will/would/may/might + NOT + a negative sentiment word
Neutral or diminished negative
Should/must + NOT+ a positive/negative sentiment word
Positive, negative, or neutral
Will (NOT)+ have p.p+ a positive/negative sentiment word
Neutral
Can/may/might/must (NOT) + have p.p + a positive sentiment word
Diminished positive
Can/may/might/must (NOT) + have p.p + a negative sentiment word
Diminished negative
Should/could/would (NOT) + have p.p+ a positive/negative sentiment word
Negative
Sentiment Analysis Utilizing Modal Expressions 195
Table 3 summarizes what is discussed in section 4. It presents the analysis result
of modal expressions by which sentiment values are expected to be correctly
assigned.
V. Conclusion
This research has identified modal expressions which have a significant impact on
the sentiment value of a sentence. The expressions are composed of a modal verb
with/without a particular linguistic component and a sentiment word. Together with a
sentiment word, the expressions assign, reverse, diminish or neutralize the sentiment
value. They have hardly mentioned or discussed in previous works since modalized
sentences conveys more complex meaning than the sentences without a modal.
Nevertheless, modalized sentences implying sentiment have to be examined because
of the influence of modality on evaluating sentence sentiment as this research
suggested. Sentiment analysis considering modality should be performed in order to
produce more accurate results.
Table 3 in section 4 summarizes the result of this research. It presents a sentiment
classification which is expected to contribute to improve the performance of a
sentiment analysis system. However, this classification has to be updated with further
researches on identifying more patterns to influence sentiment values. For example,
the sentiment of a modalized sentence can be either diminished or neutralized when
the predicate of the sentence is composed of a psych verb. A comparative structure
together with a modal verb can also be used as a significant linguistic cue to shift
the sentiment. Since issues with those patterns related to sentiment analysis are big
enough to become a separate research topic, they are reserved for future works.
196 영미연구 제39집
Notes
1) The underlined word refers to a sentiment word.
2) As for evaluating the sentiment in sentences, we consulted with a native English speaker who studies Linguistics. The symbols, (+), (−) and (0) represent positive, negative and neutral sentiment respectively.
3) Polarity values of a sentiment word differ depending on a domain. In the domain of small appliances or portable machines, for example, ‘big’ would be interpreted as negative while ‘small’ would be positive.
Sentiment Analysis Utilizing Modal Expressions 197
Works Cited
Ganapathibhotla, Murthy, and Bing Liu. “Mining opinions in comparative sentences.”
Proceedings of the 22nd International Conference on Computational
Linguistics. Vol 1. Association for Computational Linguistics, 2008. Print.
Hu, Minqing, and Bing Liu. “Mining and summarizing customer reviews.”
Proceedings of the tenth ACM SIGKDD international conference on
Knowledge discovery and data mining. ACM, 2004. Print.
Jindal, Nitin, and Bing Liu. “Identifying comparative sentences in text documents.”
Proceedings of the 29th annual international ACM SIGIR conference on
Research and development in information retrieval. ACM, 2006. Print.
Kim, Eunhee. Utilizing Contextual Cues for Sentiment Analysis of Movie Reviews.
Diss. Hankuk University of Foreign Studies, 2013. Dissertation.
Liu, Bing. Sentiment analysis: Mining opinions, sentiments, and emotions. Cambridge:
Cambridge UP, 2015. Print.
______, and Yang, et al. “Sentence-level sentiment analysis in the presence of
modalities.” International Conference on Intelligent Text Processing and
Computational Linguistics. Springer Berlin Heidelberg, 2014. Print.
Narayanan, Ramanathan, Bing Liu, and Alok Choudhary. “Sentiment analysis of
conditional sentences.” Proceedings of the 2009 Conference on Empirical
Methods in Natural Language Processing: Volume 1-Volume 1. Association
for Computational Linguistics, 2009. Print.
Palmer, Frank Robert. Mood and Modality. Cambridge: Cambridge UP, 2001. Print.
Pang, Bo, and Lillian Lee. “A sentimental education: Sentiment analysis using
subjectivity summarization based on minimum cuts.” Proceedings of the
42nd annual meeting on Association for Computational Linguistics.
198 영미연구 제39집
Association for Computational Linguistics, 2004. Print.
Socher, Richard, et al. “Recursive deep models for semantic compositionality over a
sentiment treebank.” Proceedings of the conference on empirical methods in
natural language processing (EMNLP) 1631 (2013). Print.
Taboada, Maite, et al. “Lexicon-based methods for sentiment analysis.”
Computational linguistics 37.2 (2011): 267-307. Print.
Wiebe, Janyce, et al. “Learning subjective language.” Computational linguistics 30.3
(2004): 277-308. Print.
Wilson, Theresa, et al. “Opinion Finder: A system for subjectivity analysis.”
Proceedings of hlt/emnlp on interactive demonstrations. Association for
Computational Linguistics, 2005. Print.
Sentiment Analysis Utilizing Modal Expressions 199
국문초록
양태 표현을 활용한 감성분석
장 지 원 (한국외국어대학교)김 지 은 (한국외국어대학교)
본 연구는 감성어를 포함하고 있는 영어 문장의 극성값에 영향을 미치는 양태
표현들을 감성분석에서 활용할 수 있도록 분석하고 분류한다. 양태 표현은 화자
가 취하는 다양한 의미론적 태도를 드러내는 문법적 요소이다. 양태 특유의 속
성상 양태화된 문장의 의미는 정확히 해석하기가 쉽지 않고 양태 표현을 통해
나타나는 감정가치 또한 정확히 분석하는 데 어려움이 있는 것이 사실이다. 그
러나 양태 표현이 문장에서 화자의 감성을 표현·해석하는 데 핵심적 방법이자
단서가 되므로 감성가치에 영향을 미치는 양태 표현을 분류하고 분석하는 작업
은 감성분석의 중요한 연구과제일 수밖에 없다. 본 연구에서는 감성분석의 정확
도를 향상시키기 위해 영화 리뷰 데이터에서 양태 표현을 포함하고 있는 문장
을 수집하여 분석하고 분류하였다. 데이터 분석을 통해 양태동사를 포함하고 있
는 표현의 유형, 문장 내의 문맥 구성요소, 계산된 극성값 등을 기준으로 양태
표현을 분류했다. 본 연구는 선행연구에서 다루지 않았던 양태 표현들을 분석했
을 뿐 아니라 이를 포함하고 있는 문장의 극성값을 판단하여 감성분석에 적용
할 수 있는 토대를 마련했다는 점에서 감성분석 시스템의 질적 향상에 기여할
것이다.
주제어: 감성분석, 주관성, 양태, 양태동사, 극성값
200 영미연구 제39집
논문접수일: 2017.01.20.심사완료일: 2017.02.06게재확정일: 2017.02.08
이름: 장지원 (제 1저자)
소속: 한국외국어대학교
주소: 서울특별시 동대문구 이문로 107 한국외국어대학교 영어대학 영어학과
이메일: [email protected]
이름: 김지은 (교신저자)
소속: 한국외국어대학교
주소: 서울특별시 동대문구 이문로 107 한국외국어대학교 영어대학 영어학과
이메일: [email protected]